Computing over Unreliable Communication Networks
Abstract:
We take the unifying view of systems interacting over communication networks as distributed computing systems and propose to study them as networked control systems. Since averaging is central operation to much science and engineering, we first study the problem of distributed averaging over unreliable networks. We point out that a popular and well-behaved algorithm can instead generate a collective global complex behavior when the inter-agent communication happens over unreliable links. To mitigate the effects of the unreliable information exchange we propose a new distributed averaging algorithm robust to noise and intermittent communication. The algorithm and the control perspective are the basis for the development of new distributed optimization systems that we can analyze and design as networked control systems. The approach applies to multi-agent cooperative applications and opens up several directions of research.